roadmap(#217): loud WARNING below xhigh — max preferred, xhigh floor by BaseInfinity · Pull Request #236 · BaseInfinity/claude-sdlc-harness

BaseInfinity · 2026-04-24T20:48:35Z

Summary

Close the coherence gap between the docs (max preferred, xhigh floor) and the hook (currently soft-nudges anything ≠ xhigh, including max — backwards).

effort=max or xhigh → silent (at or above floor)
effort=high/medium/low or unset → LOUD WARNING block with SDLC compliance mention, /effort max primary, /effort xhigh floor alternative, opus[1m] model reminder
Removed duplicate effort/model check from instructions-loaded-check.sh (single source of truth = model-effort-check.sh)

Why

hooks/model-effort-check.sh was flagging max as "upgrade available", which was the opposite of what SDLC.md / CLAUDE_CODE_SDLC_WIZARD.md already say. On Opus 4.7, high or below actively breaks SDLC compliance (shallow reasoning, skipped TDD, dropped self-review), so the hook should be louder there, not softer on max.

Codex cross-model review

Round 1: 7/10 NOT CERTIFIED (dupe nudge in instructions-loaded + stale doc refs)
Round 2: 8/10 NOT CERTIFIED (one stale line missed at CHANGELOG.md:120)
Round 3: 10/10 CERTIFIED

Test plan

2 new TDD tests (test_model_effort_check_max_is_silent, test_model_effort_check_below_xhigh_loud_warning)
1 new regression test (test_instructions_loaded_no_duplicate_effort_nudge)
Updated test_hooks_recommend_opus_1m_alias to assert single-source-of-truth
bash tests/test-hooks.sh → 119/119 pass
bash tests/test-doc-consistency.sh → 22/22 pass
bash tests/test-workflow-triggers.sh → 165/165 pass
Output size cap (500 chars) verified at 432 chars
Validate CI job green

…floor Current hook treats any effort ≠ xhigh as 'upgrade available' — including max (the preferred default), which is backwards. Docs already say 'max preferred, xhigh floor'. Close the coherence gap. Behavior (post-change): - effort=max or xhigh → silent (at or above floor) - effort=high/medium/low or unset → LOUD WARNING block: WARNING marker, SDLC compliance mention, /effort max primary recommendation, /effort xhigh floor alternative, opus[1m] model reminder Also removes a duplicate effort/model check from instructions-loaded-check.sh that used the same stale xhigh-as-recommended logic and double-printed on SessionStart. Single source of truth is now model-effort-check.sh. Tests: - 2 new TDD tests in test-hooks.sh (max silent, below-xhigh loud) - 1 new regression test (instructions-loaded doesn't re-declare the nudge) - Updated test_hooks_recommend_opus_1m_alias to assert single-source-of-truth - 119/119 hook tests pass, 22/22 doc-consistency tests pass Doc refs: - CHANGELOG.md:120+140 and ROADMAP.md:134 annotated as superseded by #217 - Live docs (SDLC.md, CLAUDE_CODE_SDLC_WIZARD.md) already say max preferred Codex cross-model review: - Round 1: 7/10 NOT CERTIFIED (dupe nudge + stale doc refs) - Round 2: 8/10 NOT CERTIFIED (one stale line missed at CHANGELOG.md:120) - Round 3: 10/10 CERTIFIED

github-actions · 2026-04-24T20:49:50Z

Claude encountered an error —— View job

I'll analyze this and get back to you.

* release: v1.37.0 Bundles two Codex-certified PRs: - PR #235 (ROADMAP #231 Phase 1): delete monthly-research.yml — 519 lines + 4 claude-code-action steps removed, zero merged artifacts in 30d, $11-23/mo API burn eliminated. Stubbed 17 test_monthly_* assertions (#231 Phase 1 pattern), marked 5 live-doc sections REMOVED/historical. Codex 3-round: 9/10. - PR #236 (ROADMAP #217): loud WARNING below xhigh effort — max preferred, xhigh floor, single source of truth. Dupe effort check removed from instructions-loaded-check.sh. 2 TDD tests + 1 regression test. Codex 3-round: 10/10. Version parity: - package.json, SDLC.md metadata + table, CLAUDE_CODE_SDLC_WIZARD.md (2 refs), .claude-plugin/plugin.json, .claude-plugin/marketplace.json → 1.37.0 - skills/update/SKILL.md Step 3 example refreshed - CHANGELOG entry "[1.37.0] - 2026-04-24" ROADMAP: - #232: /update-wizard should mimic 'claude update' UX - #233: mixed-mode coder+reviewer tier (Sonnet coder + Opus reviewer for simple repos, inspired by codex-sdlc-wizard) - #234: GitHub About populated (description, homepage, 8 topics) Release review (Codex round 1 → 2): - REL-001 fixed: commit now exists on release branch (was empty) - REL-002 fixed: '2 claude-code-action blocks' corrected to '4 steps' - REL-003 fixed: .reviews/handoff.json + untracked .claude/hooks excluded * release(v1.37.0): fix CHANGELOG line 26 count (2 → 4 claude-code-action steps) Codex round-2 recheck caught that the 'Removed' section bullet still said '2 API blocks' — line 11 was fixed, line 26 was not. Align with reality: pre-delete workflow had 4 uses: anthropics/claude-code-action@v1 steps.

…rn its keep?' pass (#313)

* bloat: BASELINE block fires once per CC session (v1.69.0) hooks/sdlc-prompt-check.sh emitted a ~250-token static SDLC BASELINE block on every UserPromptSubmit. After the SDLC skill auto-invokes, that block is duplicate context — ~12K tokens/session wasted for any user with >3 prompts. Now gated on a per-session_id sentinel under \$SDLC_WIZARD_CACHE_DIR/baseline-shown-<id>, atomic-claimed via subshell 'set -C' (noclobber) so N parallel fires emit exactly once. Pruned at 7d. SETUP-not-complete + EFFORT-bump warnings still fire every prompt. No-session_id stdin keeps current emit-every-fire behavior (legacy CC + direct shell tests). Codex round 1 caught two P1s, both fixed: (a) concurrency race in naive 'check then write-after-emit' pattern; (b) session_id extraction was jq-coupled — now uses grep+sed so the gate works without jq. Round 2 CERTIFIED 9/10. Regression tests: 50-parallel + restricted- PATH-no-jq. Discovered during ROADMAP #236 functional-bloat audit. Audit method: measure cost x frequency, judge value — not blind delete-and-see. Other hooks audited (model-effort-check, precompact-seam-check, token-spike-check) earned their keep — silent at healthy state. * ci: list test-baseline-fires-once.sh in CONTRIBUTING.md test-workflow-triggers.sh asserts CONTRIBUTING.md lists every test script that CI runs. Adding the new test was missed in the v1.69.0 commit.

hooks/tdd-pretool-check.sh emitted a ~50-token JSON nudge ('TDD CHECK: Are you writing IMPLEMENTATION before a FAILING TEST?') on every Write/Edit/MultiEdit touching src/**. After the SDLC skill auto-invokes (which already covers TDD RED/GREEN), the per-Edit nudge is duplicate context. Typical SDLC session has 10-30 src Edits = ~0.5-1.5K wasted tokens. Same atomic-noclobber sentinel pattern proven in v1.69.0 (BASELINE gate), applied to the TDD CHECK output. New sentinel at \$SDLC_WIZARD_CACHE_DIR/ tdd-shown-<session_id>. Non-src/ files still produce zero output and crucially DO NOT consume the sentinel slot (Test 6). Codex round 1 caught one P1 (new test file untracked at review time); fixed by staging before round 2. Round 2 CERTIFIED 9/10. ROADMAP #236 phase 2. Combined v1.69.0 + v1.70.0 saves ~13.5K tokens on a 50-prompt + 20-Edit session.

skills/sdlc/SKILL.md Cross-Model Review section condensed from ~70 lines to ~20 lines. Saves ~427 tokens per SDLC skill auto-invoke (4995 → 4568 tokens). The skill auto-loads on virtually every productive implement/fix/refactor task, so this is real per-session cost. What stayed: decision-making (when to run/skip, prereqs), 4-step protocol summary (preflight → handoff → reviewer → dialogue), required JSON keys including the test-asserted '"mission"', '"success"', '"failure"', '"verification_checklist"', '"pr_number":' (PreCompact self-heal opt-in #209), convergence rule, ### Release Review Focus subsection with Version parity, multi-reviewer/non-code-domain one-liner. What moved to wizard doc canonical: full handoff JSON example, full codex command example. New subsections added in CLAUDE_CODE_SDLC_WIZARD.md → Cross-Model Review Loop: #### Anti-patterns, #### Multiple reviewers, #### Non-code domains. The SKILL.md pointer now reaches all dropped content. Codex review: round 1 caught 3 P0 test-self-update.sh failures (4/10); round 2 caught 1 more P0 in test-hooks.sh asserting '"pr_number":' with colon, not bare quotes (6/10); round 3 CERTIFIED 10/10. ROADMAP #236 phase 3. Combined v1.69 + v1.70 + v1.71 saves ~14K tokens/session for typical workloads.

…floor (#236) Current hook treats any effort ≠ xhigh as 'upgrade available' — including max (the preferred default), which is backwards. Docs already say 'max preferred, xhigh floor'. Close the coherence gap. Behavior (post-change): - effort=max or xhigh → silent (at or above floor) - effort=high/medium/low or unset → LOUD WARNING block: WARNING marker, SDLC compliance mention, /effort max primary recommendation, /effort xhigh floor alternative, opus[1m] model reminder Also removes a duplicate effort/model check from instructions-loaded-check.sh that used the same stale xhigh-as-recommended logic and double-printed on SessionStart. Single source of truth is now model-effort-check.sh. Tests: - 2 new TDD tests in test-hooks.sh (max silent, below-xhigh loud) - 1 new regression test (instructions-loaded doesn't re-declare the nudge) - Updated test_hooks_recommend_opus_1m_alias to assert single-source-of-truth - 119/119 hook tests pass, 22/22 doc-consistency tests pass Doc refs: - CHANGELOG.md:120+140 and ROADMAP.md:134 annotated as superseded by #217 - Live docs (SDLC.md, CLAUDE_CODE_SDLC_WIZARD.md) already say max preferred Codex cross-model review: - Round 1: 7/10 NOT CERTIFIED (dupe nudge + stale doc refs) - Round 2: 8/10 NOT CERTIFIED (one stale line missed at CHANGELOG.md:120) - Round 3: 10/10 CERTIFIED

* release: v1.37.0 Bundles two Codex-certified PRs: - PR #235 (ROADMAP #231 Phase 1): delete monthly-research.yml — 519 lines + 4 claude-code-action steps removed, zero merged artifacts in 30d, $11-23/mo API burn eliminated. Stubbed 17 test_monthly_* assertions (#231 Phase 1 pattern), marked 5 live-doc sections REMOVED/historical. Codex 3-round: 9/10. - PR #236 (ROADMAP #217): loud WARNING below xhigh effort — max preferred, xhigh floor, single source of truth. Dupe effort check removed from instructions-loaded-check.sh. 2 TDD tests + 1 regression test. Codex 3-round: 10/10. Version parity: - package.json, SDLC.md metadata + table, CLAUDE_CODE_SDLC_WIZARD.md (2 refs), .claude-plugin/plugin.json, .claude-plugin/marketplace.json → 1.37.0 - skills/update/SKILL.md Step 3 example refreshed - CHANGELOG entry "[1.37.0] - 2026-04-24" ROADMAP: - #232: /update-wizard should mimic 'claude update' UX - #233: mixed-mode coder+reviewer tier (Sonnet coder + Opus reviewer for simple repos, inspired by codex-sdlc-wizard) - #234: GitHub About populated (description, homepage, 8 topics) Release review (Codex round 1 → 2): - REL-001 fixed: commit now exists on release branch (was empty) - REL-002 fixed: '2 claude-code-action blocks' corrected to '4 steps' - REL-003 fixed: .reviews/handoff.json + untracked .claude/hooks excluded * release(v1.37.0): fix CHANGELOG line 26 count (2 → 4 claude-code-action steps) Codex round-2 recheck caught that the 'Removed' section bullet still said '2 API blocks' — line 11 was fixed, line 26 was not. Align with reality: pre-delete workflow had 4 uses: anthropics/claude-code-action@v1 steps.

…rn its keep?' pass (#313)

* bloat: BASELINE block fires once per CC session (v1.69.0) hooks/sdlc-prompt-check.sh emitted a ~250-token static SDLC BASELINE block on every UserPromptSubmit. After the SDLC skill auto-invokes, that block is duplicate context — ~12K tokens/session wasted for any user with >3 prompts. Now gated on a per-session_id sentinel under \$SDLC_WIZARD_CACHE_DIR/baseline-shown-<id>, atomic-claimed via subshell 'set -C' (noclobber) so N parallel fires emit exactly once. Pruned at 7d. SETUP-not-complete + EFFORT-bump warnings still fire every prompt. No-session_id stdin keeps current emit-every-fire behavior (legacy CC + direct shell tests). Codex round 1 caught two P1s, both fixed: (a) concurrency race in naive 'check then write-after-emit' pattern; (b) session_id extraction was jq-coupled — now uses grep+sed so the gate works without jq. Round 2 CERTIFIED 9/10. Regression tests: 50-parallel + restricted- PATH-no-jq. Discovered during ROADMAP #236 functional-bloat audit. Audit method: measure cost x frequency, judge value — not blind delete-and-see. Other hooks audited (model-effort-check, precompact-seam-check, token-spike-check) earned their keep — silent at healthy state. * ci: list test-baseline-fires-once.sh in CONTRIBUTING.md test-workflow-triggers.sh asserts CONTRIBUTING.md lists every test script that CI runs. Adding the new test was missed in the v1.69.0 commit.

hooks/tdd-pretool-check.sh emitted a ~50-token JSON nudge ('TDD CHECK: Are you writing IMPLEMENTATION before a FAILING TEST?') on every Write/Edit/MultiEdit touching src/**. After the SDLC skill auto-invokes (which already covers TDD RED/GREEN), the per-Edit nudge is duplicate context. Typical SDLC session has 10-30 src Edits = ~0.5-1.5K wasted tokens. Same atomic-noclobber sentinel pattern proven in v1.69.0 (BASELINE gate), applied to the TDD CHECK output. New sentinel at \$SDLC_WIZARD_CACHE_DIR/ tdd-shown-<session_id>. Non-src/ files still produce zero output and crucially DO NOT consume the sentinel slot (Test 6). Codex round 1 caught one P1 (new test file untracked at review time); fixed by staging before round 2. Round 2 CERTIFIED 9/10. ROADMAP #236 phase 2. Combined v1.69.0 + v1.70.0 saves ~13.5K tokens on a 50-prompt + 20-Edit session.

skills/sdlc/SKILL.md Cross-Model Review section condensed from ~70 lines to ~20 lines. Saves ~427 tokens per SDLC skill auto-invoke (4995 → 4568 tokens). The skill auto-loads on virtually every productive implement/fix/refactor task, so this is real per-session cost. What stayed: decision-making (when to run/skip, prereqs), 4-step protocol summary (preflight → handoff → reviewer → dialogue), required JSON keys including the test-asserted '"mission"', '"success"', '"failure"', '"verification_checklist"', '"pr_number":' (PreCompact self-heal opt-in #209), convergence rule, ### Release Review Focus subsection with Version parity, multi-reviewer/non-code-domain one-liner. What moved to wizard doc canonical: full handoff JSON example, full codex command example. New subsections added in CLAUDE_CODE_SDLC_WIZARD.md → Cross-Model Review Loop: #### Anti-patterns, #### Multiple reviewers, #### Non-code domains. The SKILL.md pointer now reaches all dropped content. Codex review: round 1 caught 3 P0 test-self-update.sh failures (4/10); round 2 caught 1 more P0 in test-hooks.sh asserting '"pr_number":' with colon, not bare quotes (6/10); round 3 CERTIFIED 10/10. ROADMAP #236 phase 3. Combined v1.69 + v1.70 + v1.71 saves ~14K tokens/session for typical workloads.

85 lines, 0 references anywhere in the repo (no test, workflow, doc, or other script invokes it). Sat ~3 months on ROADMAP Back Burner with explicit gate 'Research if users care before shipping' — no demand signal materialized. Not in package.json files: list, so never shipped to npm consumers. Pure repo dead weight. Discovered via #236 method: for each scripts/*.sh, count refs in *.sh, *.yml, *.json, *.md, *.py. Threshold 0 = orphan. Six scripts inventoried; one orphan found. ROADMAP #236 marked phase-partial DONE: 3 token-amplifier fixes (v1.69.0/1.70.0/1.71.0) + this orphan delete. Categories not exercised this round (tests redundancy, CLI templates, ROADMAP archive) re-trigger on maintainer call-out. Reversible: git history retains the script if demand surfaces.

…h) (#320) scripts/audit-session-load.sh scanned only root skills/*/SKILL.md. But cli/init.js:32-35 copies SKILL.md files to .claude/skills/<name>/ SKILL.md at install time, so the audit was silently invisible to bloat in real consumer projects. Mirrors the existing .claude/hooks/ pattern. Caught by Codex strategic review on the bloat-sweep continuation question — it pointed at the asymmetry between hooks (dual-path scanned) and skills (root-only). Implementation plan went through a second Codex round for sign-off; ships with all 4 deltas applied: - inventory comment block updated to mention .claude/skills/ - new test uses existing make_temp + make_file_with_size helpers - new test wired into the explicit-call runner block - Python predicate uses path.endswith() not substring match Will double-count in dogfood (where .claude/skills/ symlinks back to skills/), same trade-off as the existing hooks scan accepts. No version bump: scripts/ + tests/ are not in package.json files list, so this never reaches npm consumers.

The .reviews/ tree is gitignored but several artifacts were committed before the gitignore line landed. They're handoff/preflight/round-N review files for now-merged PRs and hold no ongoing reference value. plans/CATCHUP.md captured the v2.1.15 → v2.1.81 catch-up in March 2026 — historical context lives in CHANGELOG (v1.8.0 entry at line 1415); the plan doc is dead weight. Kept everything that's still load-bearing: - .reviews/research-95/97/99/206/235.md (cited from ROADMAP rows) - .reviews/experiment-tracking.md (asserted by tests/test-workflow-triggers.sh:2189) - plans/AUTO_SELF_UPDATE.md (still annotated with #231 phase notes) Tests: hooks 156/156, cli 88/88, workflow 176/176, docs 35/35 — all green post-deletion (no test referenced any deleted file).

Two PRs roll up: 1. PR #330 — `hooks/precompact-seam-check.sh` no longer false-positive HOLDs `/compact` on stale `.git/REBASE_HEAD` (left behind after a clean rebase). Authoritative signal is `rebase-{merge,apply}/` dir; `REBASE_HEAD` is just a ref to the stopped/replayed commit. Hit live 2026-05-05; required `rm .git/REBASE_HEAD` workaround. Codex round 1 CERTIFIED 9/10 (one P2 comment-accuracy nit fixed). 2. PR #331 — gc(#236): -460 LOC across 15 stale review/plan artifacts committed before `.reviews/` made it into `.gitignore`. All for now-merged work; load-bearing files (research-*.md, experiment- tracking.md, AUTO_SELF_UPDATE.md) preserved. Version bumps: package.json, plugin.json, marketplace.json, SDLC.md, CLAUDE_CODE_SDLC_WIZARD.md (×2), skills/update/SKILL.md. skills/update/SKILL.md trimmed (consolidated 1.69.0–1.71.0 token-bloat phases into a single line) to keep the file under the 5K-token dogfood threshold after adding the 1.73.0 entry. Tests: cli 88/88, hooks 156/156, self-update 76/76, docs 35/35, audit 10/10. audit-session-load reports 0 trim candidates.

`.reviews/response.json` is per-cycle review scratch documented in the SDLC skill's cross-model review protocol — the file the next round writes its FIXED/DISPUTED/ACCEPTED responses to. The committed copy was leftover from gc-benchmark-workflows-001 (PR #324, merged). `.reviews/` is gitignored at directory level; this file slipped through because it was committed before the gitignore was added. No live consumer: the only test referencing the filename (tests/test-self-update.sh:400) just asserts the wizard doc describes the protocol, not that the file exists. Tests: self-update + doc-consistency green post-deletion.

Two PRs roll up: 1. PR #330 — `hooks/precompact-seam-check.sh` no longer false-positive HOLDs `/compact` on stale `.git/REBASE_HEAD` (left behind after a clean rebase). Authoritative signal is `rebase-{merge,apply}/` dir; `REBASE_HEAD` is just a ref to the stopped/replayed commit. Hit live 2026-05-05; required `rm .git/REBASE_HEAD` workaround. Codex round 1 CERTIFIED 9/10 (one P2 comment-accuracy nit fixed). 2. PR #331 — gc(#236): -460 LOC across 15 stale review/plan artifacts committed before `.reviews/` made it into `.gitignore`. All for now-merged work; load-bearing files (research-*.md, experiment- tracking.md, AUTO_SELF_UPDATE.md) preserved. Version bumps: package.json, plugin.json, marketplace.json, SDLC.md, CLAUDE_CODE_SDLC_WIZARD.md (×2), skills/update/SKILL.md. skills/update/SKILL.md trimmed (consolidated 1.69.0–1.71.0 token-bloat phases into a single line) to keep the file under the 5K-token dogfood threshold after adding the 1.73.0 entry. Tests: cli 88/88, hooks 156/156, self-update 76/76, docs 35/35, audit 10/10. audit-session-load reports 0 trim candidates.

`.reviews/response.json` is per-cycle review scratch documented in the SDLC skill's cross-model review protocol — the file the next round writes its FIXED/DISPUTED/ACCEPTED responses to. The committed copy was leftover from gc-benchmark-workflows-001 (PR #324, merged). `.reviews/` is gitignored at directory level; this file slipped through because it was committed before the gitignore was added. No live consumer: the only test referencing the filename (tests/test-self-update.sh:400) just asserts the wizard doc describes the protocol, not that the file exists. Tests: self-update + doc-consistency green post-deletion.

Rows #78, #79, #80, #215, #217 each say in their own cell text: Stale duplicate of <canonical row>; <work> shipped. v1.62.0 paperwork close. The canonical entries live earlier in the file. Keeping the duplicates just inflates the unprioritized table without adding info — every field reduces to "see <other row>." Removing them shrinks the unprioritized table by 5 rows and removes the "wait, is this still open?" friction when scanning. doc-consistency tests 35/35 still green (no test asserts row count).

The .reviews/ tree is gitignored but several artifacts were committed before the gitignore line landed. They're handoff/preflight/round-N review files for now-merged PRs and hold no ongoing reference value. plans/CATCHUP.md captured the v2.1.15 → v2.1.81 catch-up in March 2026 — historical context lives in CHANGELOG (v1.8.0 entry at line 1415); the plan doc is dead weight. Kept everything that's still load-bearing: - .reviews/research-95/97/99/206/235.md (cited from ROADMAP rows) - .reviews/experiment-tracking.md (asserted by tests/test-workflow-triggers.sh:2189) - plans/AUTO_SELF_UPDATE.md (still annotated with #231 phase notes) Tests: hooks 156/156, cli 88/88, workflow 176/176, docs 35/35 — all green post-deletion (no test referenced any deleted file).

85 lines, 0 references anywhere in the repo (no test, workflow, doc, or other script invokes it). Sat ~3 months on ROADMAP Back Burner with explicit gate 'Research if users care before shipping' — no demand signal materialized. Not in package.json files: list, so never shipped to npm consumers. Pure repo dead weight. Discovered via #236 method: for each scripts/*.sh, count refs in *.sh, *.yml, *.json, *.md, *.py. Threshold 0 = orphan. Six scripts inventoried; one orphan found. ROADMAP #236 marked phase-partial DONE: 3 token-amplifier fixes (v1.69.0/1.70.0/1.71.0) + this orphan delete. Categories not exercised this round (tests redundancy, CLI templates, ROADMAP archive) re-trigger on maintainer call-out. Reversible: git history retains the script if demand surfaces.

scripts/audit-session-load.sh scanned only root skills/*/SKILL.md. But cli/init.js:32-35 copies SKILL.md files to .claude/skills/<name>/ SKILL.md at install time, so the audit was silently invisible to bloat in real consumer projects. Mirrors the existing .claude/hooks/ pattern. Caught by Codex strategic review on the bloat-sweep continuation question — it pointed at the asymmetry between hooks (dual-path scanned) and skills (root-only). Implementation plan went through a second Codex round for sign-off; ships with all 4 deltas applied: - inventory comment block updated to mention .claude/skills/ - new test uses existing make_temp + make_file_with_size helpers - new test wired into the explicit-call runner block - Python predicate uses path.endswith() not substring match Will double-count in dogfood (where .claude/skills/ symlinks back to skills/), same trade-off as the existing hooks scan accepts. No version bump: scripts/ + tests/ are not in package.json files list, so this never reaches npm consumers.

Adds a tight /goal section to skills/sdlc/SKILL.md covering CC's native goal-mode primitive (v2.1.139+, requires v2.1.143+ for the subagent-race fix). Composes with existing wizard hooks rather than replacing them. Per the corrected research at .reviews/347-goal-mode-research-CORRECTED.md, the wrapper carries 5 load-bearing elements (each enforced by the new quality test): 1. Pre-flight checklist — workspace trusted, hooks not disabled at any settings layer, CC version floor. 2. Condition-writing guidance treating the /goal arg as an SDLC contract (measurable end state + check + constraints + hard turn/time bound). 3. Hook composition note — UserPromptSubmit/SessionStart/PreCompact fire normally per turn, so sdlc-prompt-check.sh + precompact-seam-check.sh keep gating. 4. Resume caveat — --resume restores the condition but resets counters. 5. Anti-pattern callout — the evaluator can't call tools, so /goal doesn't work for 'doneness' that lives off-transcript. Quality test (tests/test-doc-consistency.sh) greps for each of the 5 elements by keyword; existence-only would let drift quietly remove the safety guidance. Section trimmed from ~1782 → ~1100 chars to keep the skill within budget (was 4995 tokens pre-edit, now 5178 — slightly over the #236 5K trim threshold but within the 25K combined-skill load budget; new content is load-bearing and justified). 40/40 doc-consistency tests pass. No new skill, no new hook, no template. Per Prove-It Gate absorption principle: extend the existing /sdlc skill rather than building a /goal-start /goal-handoff /goal-stop trio.

CI's tests/test-audit-session-load.sh enforces the #236 5K-token soft threshold as a hard fail. Initial /goal section was 700 chars (skill at 5018 tokens — 18 over). Tightened to 488 chars (skill at 4977 tokens — 23 under) while keeping all 5 quality-test keywords. What stays (verified by test_sdlc_skill_has_goal_wrapper): - v2.1.143+ version floor (the only version that matters for safe use) - 'trusted workspace' pre-flight - 'disableAllHooks'/'allowManagedHooksOnly' pre-flight - 'hard turn/time bound' guidance + 'stop after' example - 'cannot call tools' anti-pattern callout - '--resume resets counters' caveat What dropped (informational, not load-bearing): - v2.1.139 introduction marker (subsumed by v2.1.143+ floor) - 'survives --resume not /clear' (the counter-reset note covers it) - 'Wizard hooks fire normally per turn' (implicit composition) - 'internally a Stop hook' parenthetical (still mentioned briefly) - Verbose example with multiple AND clauses The aggressive tightening keeps the safety guidance load-bearing while respecting the bloat budget the maintainer set in #236. Audit now returns OK for skills/sdlc/SKILL.md (4977/5000 tokens).

closes #347) (#351) * docs(cc-features): bump recommended baseline + document /goal, /code-review --comment, /usage, /context all CC v2.1.139 → v2.1.150 shipped 4 wizard-relevant features that the broken auto-PR pipeline (gutted in #231 Phase 3d, see #350) failed to surface: - Native /goal command (v2.1.139) — set completion condition, evaluator pass per turn until met. Survives --resume, not /clear. No disk writes. Subagent race fixed in v2.1.143. - /code-review --comment flag (v2.1.147, renamed from /simplify) — posts findings as inline GitHub PR comments. Streamlines our CI shepherd flow which currently uses manual gh pr review comments. - /usage per-category breakdown (v2.1.149) — skills/subagents/plugins/ MCP cost split. Complements scripts/audit-session-load.sh (#236). - /context all per-skill per-model token estimates (v2.1.139) — same. SDLC.md splits the existing Claude Code Baseline row into Minimum (unchanged, v2.1.111+/v2.1.105+) and Recommended (v2.1.150+) so consumers know what each version unlocks without breaking back-compat. CLAUDE_CODE_SDLC_WIZARD.md adds 4 rows to the Complementary Native Skills table with usage guidance and the same caveats we apply to /insights (qualitative-only, evaluator-can't-call-tools for /goal). Full inventory + triage of all 32 CC versions since v2.1.118 lives at .reviews/cc-feature-inventory-2026-05-24.md (gitignored). HIGH/MEDIUM/ LOW relevance breakdown + recommended adoption sequence for the follow-up PRs (hook $CLAUDE_EFFORT cleanup, #350 cadence workflow build, #347 /goal /sdlc skill wrapper). Zero code changes — pure documentation update so consumers benefit immediately even before the deeper hook/skill integrations land. * feat(sdlc): add /goal wrapper to /sdlc skill (closes #347) Adds a tight /goal section to skills/sdlc/SKILL.md covering CC's native goal-mode primitive (v2.1.139+, requires v2.1.143+ for the subagent-race fix). Composes with existing wizard hooks rather than replacing them. Per the corrected research at .reviews/347-goal-mode-research-CORRECTED.md, the wrapper carries 5 load-bearing elements (each enforced by the new quality test): 1. Pre-flight checklist — workspace trusted, hooks not disabled at any settings layer, CC version floor. 2. Condition-writing guidance treating the /goal arg as an SDLC contract (measurable end state + check + constraints + hard turn/time bound). 3. Hook composition note — UserPromptSubmit/SessionStart/PreCompact fire normally per turn, so sdlc-prompt-check.sh + precompact-seam-check.sh keep gating. 4. Resume caveat — --resume restores the condition but resets counters. 5. Anti-pattern callout — the evaluator can't call tools, so /goal doesn't work for 'doneness' that lives off-transcript. Quality test (tests/test-doc-consistency.sh) greps for each of the 5 elements by keyword; existence-only would let drift quietly remove the safety guidance. Section trimmed from ~1782 → ~1100 chars to keep the skill within budget (was 4995 tokens pre-edit, now 5178 — slightly over the #236 5K trim threshold but within the 25K combined-skill load budget; new content is load-bearing and justified). 40/40 doc-consistency tests pass. No new skill, no new hook, no template. Per Prove-It Gate absorption principle: extend the existing /sdlc skill rather than building a /goal-start /goal-handoff /goal-stop trio. * fix(sdlc): tighten /goal section to stay under #236 5K-token cap CI's tests/test-audit-session-load.sh enforces the #236 5K-token soft threshold as a hard fail. Initial /goal section was 700 chars (skill at 5018 tokens — 18 over). Tightened to 488 chars (skill at 4977 tokens — 23 under) while keeping all 5 quality-test keywords. What stays (verified by test_sdlc_skill_has_goal_wrapper): - v2.1.143+ version floor (the only version that matters for safe use) - 'trusted workspace' pre-flight - 'disableAllHooks'/'allowManagedHooksOnly' pre-flight - 'hard turn/time bound' guidance + 'stop after' example - 'cannot call tools' anti-pattern callout - '--resume resets counters' caveat What dropped (informational, not load-bearing): - v2.1.139 introduction marker (subsumed by v2.1.143+ floor) - 'survives --resume not /clear' (the counter-reset note covers it) - 'Wizard hooks fire normally per turn' (implicit composition) - 'internally a Stop hook' parenthetical (still mentioned briefly) - Verbose example with multiple AND clauses The aggressive tightening keeps the safety guidance load-bearing while respecting the bloat budget the maintainer set in #236. Audit now returns OK for skills/sdlc/SKILL.md (4977/5000 tokens).

…entries to stay under #236 5K cap CI caught stale 'Latest: 1.75.1' example in skills/update/SKILL.md which is grep-tested. Added new [1.76.0] entry summarizing this release. Tightened the [1.74.0] and [1.73.0] entries to compensate so the skill stays under the #236 5K-token bloat threshold (now 4989/5000).

…MAP cleanup (#352) * release: v1.76.0 — native /goal wrapper + CC v2.1.150 feature adoption + ROADMAP cleanup Versioned cut of 5 commits since v1.75.1. Headline: native /goal wrapper in /sdlc skill closes #347; full CC v2.1.119 → v2.1.150 feature inventory done with HIGH-relevance docs adopted; ROADMAP cleaned with new demand-signal-first entry gate (4 excise, 4 kill); #347 corrected from wrong 'no native primitive' research; #350 added to capture the cadence gap that let /goal slip 5 weeks. Full notes in CHANGELOG.md [1.76.0]. Trusted Publishing (OIDC) is the release channel — release.yml fires on tag push, npm publish runs as the workflow identity, no token to rotate. * release: v1.76.0 — also bump SDLC.md version metadata in wizard doc template CI caught a missed version-bump in the SDLC.md example template inside CLAUDE_CODE_SDLC_WIZARD.md (line 2984). tests/test-hooks.sh's test_sdlc_version_matches_wizard greps both files for the meta-comment and asserts they agree; the v1.76.0 cut updated SDLC.md itself but not the wizard doc's embedded template. Self-review lesson: when version-bumping, grep for ALL SDLC Wizard Version meta-comments across the repo, not just the canonical SDLC.md. * release: v1.76.0 — bump update skill changelog example + tighten old entries to stay under #236 5K cap CI caught stale 'Latest: 1.75.1' example in skills/update/SKILL.md which is grep-tested. Added new [1.76.0] entry summarizing this release. Tightened the [1.74.0] and [1.73.0] entries to compensate so the skill stays under the #236 5K-token bloat threshold (now 4989/5000).

…DLC binding (PR-D) Native /goal is now table-stakes across CC (v2.1.139+), Codex CLI, and likely others. Without SDLC discipline baked into the goal CONDITION itself, the Haiku evaluator rubber-stamps 'did the agent flail for 20 turns' instead of 'is the goal met correctly.' Two new load-bearing gates in skills/sdlc/SKILL.md ## Long-Running Goals section: 1. **Confidence gate — NEVER invoke below HIGH 95%.** Mirrors existing Confidence Check (plan first if below). Below 95% the evaluator has no anchor for 'is this correct'; only for 'did the agent stop.' 2. **DLC binding — condition MUST name the active DLC** (/sdlc for code, /gdlc for games, /ldlc for legal, etc.). Anchors the evaluator on 'doing it right,' not just 'doing it.' Example: /goal 'tests pass + clean tree following /sdlc, stop after 20 turns'. Quality test extended (tests/test-doc-consistency.sh::test_sdlc_skill_has_goal_wrapper): adds keyword greps for '95% confidence' / 'HIGH 95%' / 'confidence gate' AND 'DLC binding' / 'name the DLC' / 'condition MUST name'. Cross-cutting compensating trims to stay under #236 5K-token cap (skill at 4989/5000): - /goal section condensed example wording ('tests pass + clean tree' vs 'npm test=0 AND git clean') - Cross-Model Review's Multi-reviewer paragraph collapsed into one line 40/40 doc-consistency green, 10/10 audit green.

…DLC binding (PR-D) (#355) Native /goal is now table-stakes across CC (v2.1.139+), Codex CLI, and likely others. Without SDLC discipline baked into the goal CONDITION itself, the Haiku evaluator rubber-stamps 'did the agent flail for 20 turns' instead of 'is the goal met correctly.' Two new load-bearing gates in skills/sdlc/SKILL.md ## Long-Running Goals section: 1. **Confidence gate — NEVER invoke below HIGH 95%.** Mirrors existing Confidence Check (plan first if below). Below 95% the evaluator has no anchor for 'is this correct'; only for 'did the agent stop.' 2. **DLC binding — condition MUST name the active DLC** (/sdlc for code, /gdlc for games, /ldlc for legal, etc.). Anchors the evaluator on 'doing it right,' not just 'doing it.' Example: /goal 'tests pass + clean tree following /sdlc, stop after 20 turns'. Quality test extended (tests/test-doc-consistency.sh::test_sdlc_skill_has_goal_wrapper): adds keyword greps for '95% confidence' / 'HIGH 95%' / 'confidence gate' AND 'DLC binding' / 'name the DLC' / 'condition MUST name'. Cross-cutting compensating trims to stay under #236 5K-token cap (skill at 4989/5000): - /goal section condensed example wording ('tests pass + clean tree' vs 'npm test=0 AND git clean') - Cross-Model Review's Multi-reviewer paragraph collapsed into one line 40/40 doc-consistency green, 10/10 audit green.

PR #355 (v1.77.0) added skill-text guidance that `/goal` must state HIGH 95% confidence and bind to a DLC (`/sdlc`, `/gdlc`, `/ldlc`) before firing. The guidance worked at the meta level but had no runtime enforcement — exactly the failure mode the discipline gate documented (text-only guidance doesn't survive the failure mode it's trying to prevent). This adds `hooks/goal-confidence-check.sh` as a UserPromptSubmit hook that: 1. Matches `/goal <condition>` prompts (silent on `/goal` status + `/goal clear`). 2. Reads `transcript_path` from hook input (verified available on UserPromptSubmit per Anthropic hook docs at code.claude.com/docs/en/hooks), walks the JSONL to find the last assistant text message, and scans for HIGH-95% confidence patterns (`HIGH (95%`, `Confidence: HIGH`, `HIGH 95%`, etc.). 3. Greps the goal condition for a DLC binding (`/[a-z]+dlc`). 4. Emits LOUD warnings on either gap — non-blocking soft nudge (exit 0), same pattern as `model-effort-check.sh`. Registered in both channels: `cli/templates/settings.json` (npm/CLI) and `hooks/hooks.json` (plugin via ${CLAUDE_PLUGIN_ROOT}). Dedupes via `_find-sdlc-root.sh` helper so dual installs don't double-fire. Bundled paperwork: - `ROADMAP.md` adds the missing `## Research Parking Lot` section that the Demand-Signal-First entry gate references but was never created. Includes the maintenance rule (prune expired rows during quarterly triage). - `skills/sdlc/SKILL.md` adds a one-line enforcement cross-reference in the Long-Running Goals section. Trimmed adjacent prose to stay under the 5K token cap (#236). #359 triage (companion close): Of the 10 API changelog entries since 2026-04-16: 8 are clearly no-op for the wizard (MCP tunnels, SEC web search data, cache diagnostics, fast mode, AWS hosting, multiagent sessions, rate limits API, managed-agent memory — all orthogonal to SDLC enforcement). 2 are deprecations — verified zero usage in this repo: `context-1m-2025-08-07` (Sonnet 4.5/4 1M beta retired 2026-04-30; we use the separate `opus[1m]` alias) and `claude-3-haiku- 20240307` (Haiku 3 retired 2026-04-20; we never used it). #359 will be closed with the triage comment after merge. Verified locally: test-hooks 160/160 (4 new #360 tests + 156 existing), test-audit-session-load 10/10, test-cli 91/91, test-plugin all green, test-doc-consistency 41/41, test-self-update 153/153, test-setup-path 83/83.

…440) Fable-led audit + Prove-It (TDD mutation-tested) pass across the three categories ROADMAP #236 flagged as not yet exercised: (b) hook noise: fixed codex-review-stop-check.sh delivering its Stop-hook nudge via stderr (invisible on exit 0 per CC hook-delivery semantics); fixed tdd-pretool-check.sh's hardcoded /src/ pattern missing this repo's own hooks/cli/.github/workflows source dirs via a new SDLC_TDD_SRC_PATTERN env override (keeps the CLI-distributed template generic); deleted goal-confidence-check.sh entirely (no cited incident, unproven value); trimmed precompact-seam-check.sh 256->101 lines (removed the untested .reviews/handoff.json self-heal branch); silenced model-effort-check.sh's unset-effort nagging; removed sdlc-prompt-check.sh's unfired #195 effort-bump detector; added 24h-cached CC-version checking to instructions-loaded-check.sh. (c) test redundancy: consolidated overlapping assertions across 10 test files; deleted tests/test-degradation-detection.sh outright (428 lines, fully superseded, 5 tests relocated into test-doc-consistency.sh); found and fixed a latent bug as a byproduct — test_ci_max_turns_sufficient in test-workflow-triggers.sh was defined twice, and the original (shadowed) definition was vacuous, always passing regardless of ci.yml content. (f) ROADMAP archive cut: split ROADMAP.md (326 lines / 182 rows) into a trimmed active table plus new ROADMAP_ARCHIVE.md holding 85 fully-resolved rows and 5 fully-archived release-history sections; added 3 row-completeness regression tests to test-doc-consistency.sh (mutation-verified) proving no row was silently dropped or duplicated. Cross-model reviewed (Codex xhigh, review_id 236-bloat-hunt-bcf-2026-07-06): round 1 found a real gate bypass in codex-gate-check.sh (a quoted flag value containing a space, e.g. git -c user.name="A B" commit, broke the git/commit structural match) — fixed and mutation-verified; round 2 CERTIFIED 10/10. All 58 test files pass.

Maintainer asked why context was at 93% and whether quality degrades. It does, and this session's last 15% is the evidence: a '41/41 passing' where 10 tests ran after the exit block, a newline hole introduced while fixing a substring hole, a 10-minute spin from a forgotten sandbox flag, and #479 — where the correct diagnosis sat in a committed ROADMAP row hours before the wrong fix shipped. Distinct from #476 (what to load) and #220/#236 (file sizes). This is session hygiene: how a driver notices its own degradation. Scope says do NOT ship a threshold number without a source, and favors symptom-based self-checks over a percentage — plus a note that a documented rule alone failed to change behavior twice tonight.

… ladder (#472) * release: v1.88.0 — Opus 5 default driver, autocompact fix, escalation ladder Ships 8 consumer-path commits stranded since v1.87.0 (2026-07-14). **Opus 5 becomes Setup A** at `xhigh`, requires Claude Code v2.1.219+. Sonnet 5 moves to Setup B (`medium`) for simple/one-off work. Flagged as a trial — strong on paper, unproven by field data. **Autocompact fix, consumer-visible:** skills/setup/SKILL.md was writing CLAUDE_AUTOCOMPACT_PCT_OVERRIDE: "30" into consumer .claude/settings.json for Setup A. That figure came from the retired opus[1m] opt-in and was never re-derived. Per raw env-vars docs the variable only lowers the trigger where compaction is PROACTIVE, and a local Opus session is the docs' own counter-example. Setup A now writes no override. Corrected in 6 wizard-doc sites plus the writer, worded as a documentation gap rather than a runtime claim. Setup B's 75 is unchanged and still correct. **Escalation ladder:** both SKILL.md copies and the wizard doc said "ASK USER" for LOW/FAILED-2x/CONFUSED, skipping model escalation entirely. Now Fable → Codex xhigh → human, with the human reserved for priority, risk, scope, spend, and irreversible calls. Confidence is not authorization. 10 more direct-to-human routes removed including the CI flow diagram and the generated CLAUDE.md template. The shipped runtime hook was emitting the OLD ladder on every consumer session — fixed and regression-tested. Also: SDLC.md gains a 9-lesson post-mortem, and ROADMAP #468-479. Version bumped across all 9 tracked locations per SDLC.md's own version-bump checklist (the narrow greps have documented blind spots). Full suite: 61 files, 0 failing. * docs(roadmap): #480 — session-degradation hygiene is undocumented Maintainer asked why context was at 93% and whether quality degrades. It does, and this session's last 15% is the evidence: a '41/41 passing' where 10 tests ran after the exit block, a newline hole introduced while fixing a substring hole, a 10-minute spin from a forgotten sandbox flag, and #479 — where the correct diagnosis sat in a committed ROADMAP row hours before the wrong fix shipped. Distinct from #476 (what to load) and #220/#236 (file sizes). This is session hygiene: how a driver notices its own degradation. Scope says do NOT ship a threshold number without a source, and favors symptom-based self-checks over a percentage — plus a note that a documented rule alone failed to change behavior twice tonight.

Maintainer asked why context was at 93% and whether quality degrades. It does, and this session's last 15% is the evidence: a '41/41 passing' where 10 tests ran after the exit block, a newline hole introduced while fixing a substring hole, a 10-minute spin from a forgotten sandbox flag, and #479 — where the correct diagnosis sat in a committed ROADMAP row hours before the wrong fix shipped. Distinct from #476 (what to load) and #220/#236 (file sizes). This is session hygiene: how a driver notices its own degradation. Scope says do NOT ship a threshold number without a source, and favors symptom-based self-checks over a percentage — plus a note that a documented rule alone failed to change behavior twice tonight.

BaseInfinity merged commit 5080161 into main Apr 24, 2026
2 of 3 checks passed

BaseInfinity deleted the roadmap-217-loud-warning branch April 24, 2026 20:50

BaseInfinity mentioned this pull request Apr 24, 2026

release: v1.37.0 #237

Merged

9 tasks

BaseInfinity mentioned this pull request May 5, 2026

roadmap(#236): functional-bloat audit #313

Merged

2 tasks

BaseInfinity added a commit that referenced this pull request May 5, 2026

roadmap(#236): functional-bloat audit — recurring 'does this still ea…

2d306ca

…rn its keep?' pass (#313)

BaseInfinity mentioned this pull request May 5, 2026

bloat: BASELINE block fires once per CC session (v1.69.0) #314

Merged

7 tasks

BaseInfinity mentioned this pull request May 5, 2026

bloat: TDD CHECK nudge fires once per CC session (v1.70.0) #315

Merged

9 tasks

BaseInfinity mentioned this pull request May 5, 2026

bloat: trim SDLC SKILL.md Cross-Model Review section (v1.71.0) #316

Merged

9 tasks

BaseInfinity added a commit that referenced this pull request May 5, 2026

roadmap(#236): functional-bloat audit — recurring 'does this still ea…

192d587

…rn its keep?' pass (#313)

This was referenced May 5, 2026

bloat(#236): delete orphan scripts/audit-subagent-models.sh #317

Closed

bloat(#236): delete orphan scripts/audit-subagent-models.sh #318

Merged

This was referenced May 5, 2026

bloat(#236): audit script scans .claude/skills/ (consumer install path) #319

Closed

bloat(#236): audit script scans .claude/skills/ (consumer install path) #320

Merged

BaseInfinity mentioned this pull request May 6, 2026

release: v1.73.0 — precompact REBASE_HEAD fix + 15-file gc #332

Merged

3 tasks

BaseInfinity mentioned this pull request May 6, 2026

roadmap(#236): note v1.73.0 tracked-artifact sweep #333

Merged

BaseInfinity added a commit that referenced this pull request May 11, 2026

roadmap(#236): note v1.73.0 tracked-artifact sweep

42b6151

BaseInfinity mentioned this pull request May 25, 2026

feat(sdlc): /goal SDLC-discipline gates — 95% confidence floor + DLC binding (PR-D) #355

Merged

3 tasks

This was referenced May 31, 2026

fix(sdlc): teach cross-model review to use Bash background-mode (closes #364) #366

Merged

feat(hooks): /goal SDLC discipline gate (closes #360, refs #359) #369

Merged

BaseInfinity mentioned this pull request Jul 7, 2026

chore(sdlc): ROADMAP #236 bloat-hunt — hooks, tests, ROADMAP archive #440

Merged

5 tasks

This was referenced Aug 14, 2026

codex-gate blocks commands that only *mention* committing — it fires on prose and misses real invocations #588

Closed

fix(hooks): fire codex-gate on git commit in command position, not on prose (#588) #598

Merged

Navigation Menu

Uh oh!

roadmap(#217): loud WARNING below xhigh — max preferred, xhigh floor - #236

roadmap(#217): loud WARNING below xhigh — max preferred, xhigh floor#236
BaseInfinity merged 1 commit into
mainfrom
roadmap-217-loud-warning

BaseInfinity commented Apr 24, 2026

Uh oh!

github-actions Bot commented Apr 24, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

BaseInfinity commented Apr 24, 2026

Summary

Why

Codex cross-model review

Test plan

Uh oh!

github-actions Bot commented Apr 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

github-actions Bot commented Apr 24, 2026 •

edited

Loading